Two issues: OSC Space serves the OQC frontend; 60 active materials' anchor scores disagree with the canonical active-table recomputation

#70
by sdoygb - opened

Hi organizers — thanks for the challenge. Two issues found while submitting.

  1. The OSC Space frontend is the OQC (Open Quantum Challenge) app.

https://final-bench-osc-leaderboard.hf.space/
serves a page titled “Open Quantum Challenge” with QSIM/QEC tabs and a submit form posting board / display_name / method / predictions — the OQC client.
The backend is OSC-correct: /api/stats → {"candidates":4832,"active":1054}; /api/active returns the 1,054 active materials; the submit model is material_id / provisional_score / method_name.
Consequence: the documented “open the material's card in the Active Challenge tab” UI does not exist; the served index.html (27,299 bytes) is identical to the OQC page and its JS throws TypeError: Cannot read properties of undefined (reading 'find') at loadBoard, so the standings do not render. Submissions currently require calling /api/submit directly (that is how sdoygb / 2BrFe-1 → 55.375 was submitted).
Please deploy the OSC index.html that matches app.py.
2) For 60 of 1,054 active materials, submissions.csv (anchor verified_score) disagrees with the canonical formula recomputed from active_challenge.csv.

The documented formula I = N(E_F)·A_d(δ)·300 matches exactly for 994 materials and disagrees for 60.
Example: 1CuBr2-1 (Br2Cu). Anchor row: δ=0.1667, A_d=0.05569 → 22.012. active_challenge.csv: δ=0.7222, A_d=0.00718 → 2.838.
The leaderboard recomputes from the active table (canon in app.py), so the anchor row displays 2.838 with claimed=22.012; participants who submitted this material (H-Collection, LYJ37Fii) are displayed at 2.838 although their claim came from the published anchor.
Question: is this the intended “corrected band-filling convention” (FAQ #61 rule 5)? If so, could the anchor rows in submissions.csv be refreshed so that the published anchor, the instrument (osc_instrument.py reads active_challenge.csv) and the leaderboard all agree? Otherwise participants who trust the anchor are silently scored down by up to ~8×.
I have the full list of the 60 (material_id, both δ/A_d pairs, both scores) and can paste or send it.
Thanks!

OSC — active materials whose anchor score disagrees with the canonical active-table score

Source: FINAL-Bench/OSC-Superconductor (active_challenge.csv, submissions.csv, scoring_meta.json), checked 2026-11-03.

Canonical formula (dataset README, and canon in app.py):
I = N(E_F) * max(0, A_d(delta)) * 300, scale = 300.

It matches exactly for 994 / 1054 active materials. The 60 below disagree.

  • active score = recomputed from active_challenge.csv (what the live leaderboard displays).
  • anchor score = verified_score in submissions.csv.
  • ratio = anchor / active (inf when the active score is 0).

60 active materials: active-table score vs submissions.csv anchor score
(format: material_id formula active_score -> anchor_score)

1CHf2-1 CHf2 7.208 -> 33.225
1CuBr2-1 Br2Cu 2.838 -> 22.012
1CuCl2-2 Cl2Cu 0.000 -> 20.999
1CSc2-1 CSc2 0.000 -> 20.697
1CV2-1 CV2 1.131 -> 18.068
1TiLi2-1 Li2Ti 0.000 -> 16.341
1FeO2-2 FeO2 0.000 -> 16.244
1CuSe2-3 CuSe2 1.012 -> 16.173
2NiS-3 Ni2S2 0.000 -> 16.009
1ScS2-1 S2Sc 1.908 -> 14.802
1CrAs2-1 As2Cr 2.174 -> 14.311
1CuLi2-1 CuLi2 1.782 -> 13.822
1NiSe2-2 NiSe2 0.860 -> 13.752
2FeO-5 Fe2O2 0.000 -> 12.585
2ScSe-2 Sc2Se2 1.073 -> 12.369
1CuSe2-2 CuSe2 0.744 -> 11.884
1ScSe2-1 ScSe2 0.985 -> 11.354
1NiCl2-2 Cl2Ni 0.000 -> 10.708
1NSc2-1 NSc2 0.630 -> 10.063
2FTi-1 F2Ti2 1.289 -> 9.994
1FeSe2-3 FeSe2 2.770 -> 9.385
2FeSe-1 Fe2Se2 0.673 -> 9.145
1TiAs2-1 As2Ti 1.367 -> 9.000
2NiS-4 Ni2S2 0.780 -> 8.991
1FeS2-3 FeS2 2.479 -> 8.397
1NiH2-1 H2Ni 0.427 -> 7.835
2FeS-1 Fe2S2 0.475 -> 7.598
1NiS2-2 NiS2 0.442 -> 7.066
2NiSe-4 Ni2Se2 0.000 -> 7.029
1AsNi2Se2-1 AsNi2Se2 0.516 -> 7.015
2CuSe-5 Cu2Se2 0.000 -> 7.004
1AsHf2Te2-1 AsHf2Te2 0.422 -> 6.747
2BrTi-1 Br2Ti2 0.351 -> 6.453
1AsCo2Te2-1 AsCo2Te2 0.355 -> 5.667
2CaCl-1 Ca2Cl2 0.362 -> 4.177
1TiH2-1 H2Ti 0.281 -> 3.823
2CuSe-3 Cu2Se2 12.768 -> 3.367
2CuSe-4 Cu2Se2 0.224 -> 3.049
2GaP-1 Ga2P2 0.000 -> 2.029
1NSr2-1 NSr2 3.308 -> 1.872
1NCa2-1 Ca2N 2.883 -> 1.631
1NiSe2-1 NiSe2 0.101 -> 1.377
1VLi2-1 Li2V 11.472 -> 1.289
1NV2-1 NV2 9.740 -> 0.804
1NCr2-1 Cr2N 13.889 -> 0.756
1C2S2Nb3-1 C2Nb3S2 7.789 -> 0.643
1CrO2-2 CrO2 0.042 -> 0.568
1GeSe2-2 GeSe2 0.000 -> 0.374
1NVCl2-1 Cl2NV 0.000 -> 0.263
1CoO2-2 CoO2 13.693 -> 0.000
1CuS2-3 CuS2 22.604 -> 0.000
1CoH2-1 CoH2 8.195 -> 0.000
1VH2-1 H2V 8.107 -> 0.000
1AsSe2-1 AsSe2 8.742 -> 0.000
1PSe2-1 PSe2 7.654 -> 0.000
1AsS2-1 AsS2 7.468 -> 0.000
1CuH2-1 CuH2 2.228 -> 0.000
2AgTe-1 Ag2Te2 5.286 -> 0.000
2CuSe-2 Cu2Se2 7.875 -> 0.000
2CoSe-1 Co2Se2 16.359 -> 0.000

FINAL_Bench org

Thank you for the precise report. Both items are resolved. (1) The OSC Space serves the OSC frontend again, so the Active Challenge cards and standings render and submissions work from the page. (2) The 60 anchor rows in submissions.csv were aligned to the canonical active-table values on Nov 4, and the two files now agree for all 1,054 materials. The leaderboard scores from the active table, so rankings did not change.

Sign up or log in to comment